Description
What would you like to be added?
Add a new field to tolerations API to ensure that the PDBs are honored while deleting pods during Taint Based Evictions
Why is this needed?
The current approach of using tolerationSeconds in pod specifications has two key limitations:
- Most application developers and cluster operators lack the detailed knowledge required to set tolerationSeconds accurately, leading to guesswork and potential issues. Some clusters have had to implement validating webhooks to prevent indefinite tolerations.
- Taint-based "evictions" ultimately result in pod deletion, but some workloads require stronger disruption handling guarantees. Cluster admins and developers have had to write custom webhooks to intercept these deletion requests.
To address these challenges, we propose introducing a new field in the Tolerations API called runAsLongAsPDB (or a similar name). This new field would allow users to express their intent to run the workload as long as the Pod Disruption Budget (PDB) exists, eliminating the need for guesswork around tolerationSeconds and providing better disruption handling guarantees. This new functionality would be designed to be backward-compatible with the existing tolerationSeconds approach, allowing both new and existing workloads to benefit from the improved disruption management capabilities
Metadata
Metadata
Assignees
Labels
Type
Projects
Status